Relax. We're not going to explain this chapter like a textbook — just like a friend talking you through it, in plain, easy-to-follow terms.
Let's think about it this way for a second
Once you retrieve something, read it. If it's unrelated, drop it. If it's relevant, include just the short excerpt plus the file name. Dumping all top-5 results into the prompt just pollutes the context. The RAG course already said "retrieve, then read." This chapter is about how you plate that already-read text. Without a citation, you don't know which text it came from. Don't grab 5 books off the shelf and just dump them on the table.
Let's connect it to everyday life
The question is about phones. What comes back: 1) phone policy, 2) computer lab, 3) cafeteria. Only add #1 to the context. Format — "Source: handbook.md / phones\nText: phones must be off during class." Don't include #2 or #3. If nothing matches at all, write "not found" in the context instead of sending a prompt that pressures it to guess.
Let's try it out together
Retrieved 3 chunks
Keep #1 handbook.md / phones
Drop #2 computer-lab.md
Drop #3 cafeteria.md
Context gets only #1 + its filenameDecide which retrieved chunks to include in context — and which to leave out.5-Minute Try
Write one question and 4 fake chunks. Show which ones you'd include, which you'd drop, and how you'd write it into the context.
One Quick Heads-Up
Include an unrelated chunk, and the model will confidently cite it anyway.